资源环境科技发展态势分析平台

Land subsidence caused by groundwater extraction has numerous negative consequences, such as loss of groundwater storage and damage to infrastructure. Understanding the magnitude, timing, and locations of land subsidence, as well as the mechanisms driving it, is crucial to implementing mitigation strategies, yet the complex, nonlinear processes causing subsidence are difficult to quantify. Physical models relating groundwater flux to aquifer compaction exist, but require substantial hydrological datasets and are time consuming to calibrate. Land deformation can be measured using InSAR and GPS, but the former is computationally expensive to estimate at scale and is subject to tropospheric and ionospheric errors, and the latter leaves many temporal and spatial gaps. In this study, we apply for the first time a machine learning approach that quantifies the relationships of various widely available input data, including evapotranspiration, land use, and sediment thickness, with land subsidence. We apply this method over the Western United States, and estimate that from 2015 to 2016, ~2.0 km³/yr of groundwater storage was lost due to groundwater pumping‐induced compaction of sediments. Subsidence is concentrated in the Central Valley of California, and the state of California accounts for 75% of total subsidence in the Western US. Other significant areas of subsidence occur in cultivated regions of the Basin and Range province. This study demonstrates that widely available ancillary data can be used to estimate subsidence at a larger scale than has been previously possible.